Projects
Grouped by what they've produced, not by what I intend
Groups here are gated on output rather than effort, which is why Core work is currently empty: the research practice and the Acropolis Edge Lab get real hours every week, but they haven't produced published findings yet, so they sit in Incubating until they do. Every entry also carries an explicit status, and the statuses move — pausing or killing something deliberately is how the top of this list stays small enough to go deep on. That's discipline, not failure.
Active receives real attention nowIncubating interesting, but not scheduledDormant paused intentionallyArchived finished or deliberately abandoned
Incubating
Under way and getting real hours, but nothing published yet. Graduates to core work when it produces findings — not when it gets exciting.
Independent Research (emerging: Noesis)
Problem: Dependable intelligence in real systems is unresolved territory — and I think it's best explored through experiments, not a pitch deck.
A research practice, not a company: two guiding questions, numbered experiments, and published write-ups. Noesis is the name this may grow into if enough real work accumulates — question → experiment → finding → tool → body of work → institution, in that order.
First write-up published. No experiments run yet.
Acropolis Edge Lab
Problem: Students get taught cloud AI; almost nobody gets to build something that runs offline on the device in their hand.
A small student research lab in Indore, first cohort in 2026. Two phone-only projects — an offline Hinglish voice Q&A assistant and on-device campus computer vision — run as experiments: question, hypothesis, build, measure, finding. It is the applied testbed for both research questions.
First cohort forming; weekly on-campus sessions.
Maintained systems
Built, shipped, and kept running. Maintenance and small improvements, not new territory.
Knos Digest
Problem: Too much noise in tech news; wanted a personalized, high-signal daily digest.
A personalized Hacker News digest delivered daily via WhatsApp — sentence-transformer embeddings, author reputation tracking, deduplication, SQLite. It ran daily for months and has real users.
Shipping daily; has run for months with real users.
Writing practice
A long-running practice with its own separate territory.
State Overflow
Problem: Wanted a space to explore statistical thinking about everyday decisions.
A newsletter applying a data scientist's lens to decisions, uncertainty, and life. Its territory stays deliberately distinct from the AI research.
One protected writing block a week.
Personal systems
Instruments I built for myself. Public because the method might be useful, not because the data is interesting.
Metrics Dashboard
Problem: Wanted to observe daily habits without a productivity app.
A static heatmap dashboard fed by Obsidian frontmatter — four daily habits, week types, milestones. Observational by design; it measures consistency, not output.
Running unattended; also the routine layer under the attention practice.
Attention Experiment
Problem: A decade of interrupt-driven days cost me something in sustained technical thought, and I wanted to know how much.
A four-month personal protocol running Aug–Dec 2026 — duration, mode, depth, switches, pull, and out, tracked without a composite score. A personal practice, not an AI research program.
Running since August 2026.
Dormant & archived
Paused or finished on purpose. Keeping the list honest is what keeps the top of it small.
Memo Agent
Problem: Wanted to save and retrieve links and notes from messaging apps without leaving the conversation.
A keyword-triggered knowledge base across WhatsApp, Telegram, and Discord — automatic metadata extraction, auto-tagging, local SQLite. Paused deliberately while the research work has my mornings.